Analysis
Here are basic counts for how many beers we have in the final data by both broad style and substyle. LOTS of IPAs, although that shoudl surprise no one :) The top four substyles are all IPAs or American pales.
top_beers3 %>%
count(broad_style, sort = TRUE) %>%
head(20)## # A tibble: 20 x 2
## broad_style n
## <chr> <int>
## 1 IPA 12269
## 2 Stout 5386
## 3 Lager 3945
## 4 Pale Ale 3537
## 5 Wheat Beer 2156
## 6 Farmhouse Ale 1980
## 7 Porter 1952
## 8 Wild Ale 1564
## 9 Red Ale 1453
## 10 Pilsner 1300
## 11 Sour 1270
## 12 Brown Ale 1069
## 13 Strong Ale 1039
## 14 Fruit and Field Beer 931
## 15 Blonde Ale 859
## 16 Bock 634
## 17 Barleywine 573
## 18 Bitter 548
## 19 Tripel 487
## 20 Kölsch 430top_beers3 %>%
count(beer_style, sort = TRUE) %>%
head(20)## # A tibble: 20 x 2
## beer_style n
## <chr> <int>
## 1 IPA - American 4901
## 2 IPA - Imperial 3794
## 3 Pale Ale - American 2695
## 4 IPA - New England 2402
## 5 Stout - American Imperial 1981
## 6 Farmhouse Ale - Saison 1829
## 7 Wild Ale 1564
## 8 Red Ale - American Amber / Red 1116
## 9 Porter - American 1032
## 10 Fruit and Field Beer 931
## 11 Stout - American 875
## 12 Stout - Sweet / Milk 786
## 13 Stout - Russian Imperial 765
## 14 Pilsner - German 762
## 15 Blonde Ale - American 749
## 16 Wheat Beer - Hefeweizen 685
## 17 Wheat Beer - Witbier 662
## 18 Sour - Berliner Weisse 652
## 19 Brown Ale - American 645
## 20 Wheat Beer - American Pale 586Comparing styles
First, let’s take a look at the raw number and percent of each style that is either rated above a 4 by BA users or above a 90 in the BA score. Above a 90 indicates that a beer is either in the “Outstanding” or “World Class” categories.
## num/percent of beers > 4 rating
top_beers3 %>%
group_by(broad_style) %>%
summarize(n = n(),
n_over4 = sum(avg >= 4, na.rm = TRUE),
percent_over4 = round((n_over4 / n),2)) %>%
filter(!is.na(broad_style)) %>%
ggplot(aes(percent_over4, reorder(broad_style, percent_over4), fill = broad_style)) +
geom_col() +
theme_classic() +
theme(legend.position = "none",
text = element_text(size=9)) +
labs(y = "Beer Style", x = "Percent of Beers with an Average Rating >= 4",
title = "Percent of Beers with an Average Rating >= 4") +
scale_x_continuous(labels = scales::percent_format()) 
top_beers3 %>%
group_by(broad_style) %>%
summarize(n = n(),
n_over4 = sum(avg >= 4, na.rm = TRUE),
percent_over4 = round((n_over4 / n),2)) %>%
filter(!is.na(broad_style)) %>%
ggplot(aes(n_over4, reorder(broad_style, n_over4), fill = broad_style)) +
geom_col() +
theme_classic() +
theme(legend.position = "none",
text = element_text(size=9)) +
labs(y = "Beer Style", x = "Number of Beers with an Average Rating >= 4",
title = "Number of Beers with an Average Rating >= 4")
top_beers3 %>%
group_by(broad_style) %>%
summarize(n = n(),
n_over90 = sum(score >= 90, na.rm = TRUE),
percent_over90 = round((n_over90 / n),2)) %>%
filter(!is.na(broad_style)) %>%
ggplot(aes(percent_over90, reorder(broad_style, percent_over90), fill = broad_style)) +
geom_col() +
theme_classic() +
theme(legend.position = "none",
text = element_text(size=9)) +
labs(y = "Beer Style", x = "Percent of Beers with a BA Score >= 90",
title = "Percent of Beers by General Style with a BA Score >= 90") +
scale_x_continuous(labels = scales::percent_format())
Wild ales and lambics do well just about however you slice the data. Obviously the list of lambic brewers and blenders in the world is small, but it’s still remarkable to have over 60% of a style rated as either outstanding or world class.
The fact that the single highest number of amazing (4+ by reader ratings) IPAs speaks to the insane number of IPAs produced right now. And it’s no surprise that stouts and wild ales are the second and (distant) third-place styles – those are usually the kinds of beers we think of as hype whales.
Also of note: based on reader averages, high-ABV styles like quads, (many) stouts, barleywines, and (some) IPAs are at or near the top of the list of the most 4+ rated beers. Again, these are whale-ish beers, like Dark Lord, which costs $170 for just five bottles, Hunahpu or Pliny the Younger.
What about the distribution of reviews within a style? Using boxplots and the interquartile range, we can take a look at the most and least-divisive beer styles. The idea here is that beer ratings for some styles are more clustered together than others, while some styles have both very high and very low-rated beers.
## distribution of reviews
top_beers3 %>%
filter(!is.na(broad_style)) %>%
ggplot(aes(avg, reorder(broad_style, avg), color = broad_style)) +
#geom_jitter(aes(alpha = 0.3)) +
geom_boxplot() +
theme_classic() +
theme(legend.position = "none",
text = element_text(size=9)) +
labs(y = "Beer Style", x = "Avg Rating")
## Most divisive styles (within-style IQR)
top_beers3 %>%
group_by(broad_style) %>%
summarize(iqr = IQR(avg)) %>%
filter(!is.na(broad_style)) %>%
arrange(desc(iqr)) %>%
slice_max(order_by = iqr, n = 20) %>%
ggplot(aes(iqr, reorder(broad_style, iqr), color = broad_style)) +
geom_point() +
geom_segment(aes(x = 0, xend = iqr, y = broad_style, yend = broad_style)) +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
legend.position = "none",
text = element_text(size=9)) +
labs(x = "Interquartile range", y = "Beer Style",
title = "Most Divisive Beer Styles")
## Most divisive substyles (within-style IQR)
top_beers3 %>%
group_by(beer_style) %>%
summarize(iqr = IQR(avg)) %>%
filter(!is.na(beer_style)) %>%
arrange(desc(iqr)) %>%
slice_max(order_by = iqr, n = 20) %>%
ggplot(aes(iqr, reorder(beer_style, iqr), color = beer_style)) +
geom_point() +
geom_segment(aes(x = 0, xend = iqr, y = beer_style, yend = beer_style)) +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
text = element_text(size=9),
legend.position = "none") +
labs(x = "Interquartile range", y = "Beer substyle",
title = "Most Divisive Beer Substyles")
# Least divisive substyles (within-style IQR)
top_beers3 %>%
group_by(beer_style) %>%
summarize(iqr = IQR(avg)) %>%
filter(!is.na(beer_style)) %>%
arrange(desc(iqr)) %>%
slice_min(order_by = iqr, n = 20) %>%
ggplot(aes(iqr, reorder(beer_style, -iqr), color = beer_style)) +
geom_point() +
geom_segment(aes(x = 0, xend = iqr, y = beer_style, yend = beer_style)) +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
text = element_text(size=9),
legend.position = "none") +
labs(x = "Interquartile range", y = "Beer substyle",
title = "Least Divisive Beer Substyles") 
Some thoughts here:
- The median wild ale is likely to be pretty good. These beers, like quads, lambics and brett beers, are difficult and/or take a long time to produce. Essentially, the brewer usually has to be pretty accomplished to make and sell one of these styles, so it’s likely to be pretty good. And of course, those long production times and large amounts of raw materials mean that these beers are often more expensive than most. So, since these are consumer reviews, it’s likely that drinkers either somewhat, if unconsciously, inflate their ratings to justify the cost, or just have a higher perceived enjoyment of the beer just because the beer style and marketing signal that it should be a good beer.
- Light lagers like Bud and High Life are among the most divisive beers. Based on a previous analysis it looks like some of this variation can be explained by joke reviews. Other kinds of lagers are also divisive, including European dark lagers (like Dunkels, Schwarzbiers, and Tmave Pivos), Dortmunders, and Japanese rice lagers. I think this may be partly because even some of the best lagers have less intense flavors than something like an IPA, and partly because lagers as a broad category often get lumped together with light lagers.
- Other divisive styles, like Kvass, Lambic, and Gueuze, may be much different than anything a drinker has ever had before. They’re often sour, smell funky (with descriptors like “horse blanket” and “barnyard”), and are rare, so they’re unlike most other beer styles that people have tried before.
- But interestingly, some of those similarly-rare styles like smoked porter, braggot, brett beer, and Biere de Garde are among the least-divisive styles. In part I think this is because these styles are comparatively rare, so the sample size of available beers is lower. And I’d argue that unlike Lambic/Gueuze, they’re rare enough that many people might not get randomly offered one at some point, like a Lambic. For example, fruited Lambics are available at a lot of fancy grocery stores, and I wouldn’t think it would be uncommon for a beer geek friend to share a gueuze with people who aren’t as in to beer, because of the style’s reputation and because they’re nice for special occasions. Basically, for the least-divisive styles, the kinds of people that usually review them on Beer Advocate are also their biggest fans. You don’t typically just stumble onto a brett beer sight unseen.
The relationship between ABV and ratings
As noted above, high-ABV beer styles are often highly-rated, but what about higher-ABV beers within a style?
## Relationships between things
## ABV vs. ratings (broad style and beer style)
top_styles <- top_beers3 %>%
count(broad_style, sort = TRUE) %>%
top_n(9)
top_beers3 %>%
filter(broad_style %in% top_styles$broad_style,
abv < 20) %>%
ggplot(aes(abv, avg)) +
geom_point(aes(alpha = 0.4, color = broad_style)) +
geom_smooth() +
facet_wrap(~broad_style, scales = "free_x") +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
legend.position = "none") +
labs(y = "Avg Rating", x = "ABV", title = "ABV vs. Avg User Rating by General Style")
top_styles <- top_beers3 %>%
count(beer_style, sort = TRUE) %>%
top_n(9)
top_beers3 %>%
filter(abv < 20,
beer_style %in% top_styles$beer_style) %>%
ggplot(aes(abv, avg)) +
geom_point(aes(alpha = 0.4, color = beer_style)) +
geom_smooth() +
facet_wrap(~beer_style, scales = "free_x") +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
legend.position = "none") +
labs(y = "Avg Rating", x = "ABV", title = "ABV vs. Avg User Rating by Substyle")
In general, for most of the top-9 styles, there’s not a strong relationship between ABV and average BA user rating. However, for stouts, wheat beers, and porters, there is a small positive correlation. Interestingly, for substyles, there’s not as strong of a relationship between ABV and rating for double IPAs as for imperial stouts.
## Ratings and Avg score
top_beers3 %>%
filter(beer_style %in% top_styles$beer_style) %>%
ggplot(aes(ratings, avg)) +
geom_point(aes(alpha = 0.4, color = beer_style)) +
geom_smooth() +
facet_wrap(~beer_style, scales = "free_x") +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
legend.position = "none") +
labs(y = "Avg Rating", x = "Number of Ratings", title = "Number of Ratings vs. Avg User Rating by General Style")
## top breweries by style (brewery has to have 3 or more examples)
library(tidytext)
top_styles <- top_beers3 %>%
count(broad_style, sort = TRUE) %>%
top_n(9)
top_beers3 %>%
group_by(broad_style, brewery) %>%
summarize(n = n(),
avg_rating = round(mean(avg, na.rm = TRUE), 2)) %>%
filter(n > 2,
!is.na(broad_style),
broad_style %in% top_styles$broad_style) %>%
group_by(broad_style) %>%
slice_max(n = 10, order_by = avg_rating) %>%
ggplot(aes(avg_rating, reorder_within(brewery, avg_rating, broad_style),
color = broad_style)) +
geom_point() +
geom_segment(aes(x = 0, xend = avg_rating,
y = reorder_within(brewery, avg_rating, broad_style),
yend = reorder_within(brewery, avg_rating, broad_style))) +
facet_wrap(~broad_style, scales = "free_y") +
scale_y_reordered() +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
text = element_text(size=9),
legend.position = "none") +
labs(x = "Average Rating", y = "Brewery", title = "Top Breweries by Style")
top_styles <- top_beers3 %>%
count(beer_style, sort = TRUE) %>%
top_n(9)
top_beers3 %>%
group_by(beer_style, brewery) %>%
summarize(n = n(),
avg_rating = round(mean(avg, na.rm = TRUE), 2)) %>%
filter(n > 2,
!is.na(beer_style),
beer_style %in% top_styles$beer_style) %>%
group_by(beer_style) %>%
slice_max(n = 10, order_by = avg_rating) %>%
ggplot(aes(avg_rating, reorder_within(brewery, avg_rating, beer_style),
color = beer_style)) +
geom_point() +
geom_segment(aes(x = 0, xend = avg_rating,
y = reorder_within(brewery, avg_rating, beer_style),
yend = reorder_within(brewery, avg_rating, beer_style))) +
facet_wrap(~beer_style, scales = "free_y") +
scale_y_reordered() +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
text = element_text(size=7),
legend.position = "none") +
labs(x = "Average Rating", y = "Brewery", title = "Top Breweries by Substyle")
## Breweries that are awesome at multiple styles (best breweries in the country?)
## (number of styles that a brewery has a top 20 avg rating in)
top_beers3 %>%
group_by(broad_style, brewery) %>%
summarize(n = n(),
avg_rating = round(mean(avg, na.rm = TRUE), 2)) %>%
filter(n > 2) %>%
group_by(broad_style) %>%
slice_max(order_by = avg_rating, n = 20) %>%
ungroup() %>%
count(brewery, sort = TRUE) %>%
slice_max(order_by = n, n = 20) %>%
ggplot(aes(n, reorder(brewery,n), fill = brewery)) +
geom_col() +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
text = element_text(size=10),
legend.position = "none") +
labs(x = "Number of Styles w/ Top 20 Avg Rating", y = "Brewery",
title = "Breweries That Are Awesome at Multiple Styles")
## By score > Rating
top_beers3 %>%
group_by(broad_style, brewery) %>%
summarize(n = n(),
avg_rating = round(mean(score, na.rm = TRUE), 2)) %>%
filter(n > 2) %>%
group_by(broad_style) %>%
slice_max(order_by = avg_rating, n = 20) %>%
ungroup() %>%
count(brewery, sort = TRUE) %>%
slice_max(order_by = n, n = 20) %>%
ggplot(aes(n, reorder(brewery,n), fill = brewery)) +
geom_col() +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
text = element_text(size=10),
legend.position = "none") +
labs(x = "Number of Styles w/ Top 20 Avg Score", y = "Brewery",
title = "Breweries That Are Awesome at Multiple Styles")
top_beers3 %>%
group_by(beer_style, brewery) %>%
summarize(n = n(),
avg_rating = round(mean(avg, na.rm = TRUE), 2)) %>%
filter(n > 2) %>%
group_by(beer_style) %>%
slice_max(order_by = avg_rating, n = 10) %>%
ungroup() %>%
count(brewery, sort = TRUE) %>%
slice_max(order_by = n, n = 20) %>%
ggplot(aes(n, reorder(brewery,n), fill = brewery)) +
geom_col() +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
text = element_text(size=10),
legend.position = "none") +
labs(x = "Number of Beer Substyles w/ Top 20 Avg Rating", y = "Brewery",
title = "Breweries That Are Awesome at Multiple Substyles")
## relationship bet score and avg rating
top_styles <- top_beers3 %>%
count(broad_style, sort = TRUE) %>%
top_n(9)
top_beers3 %>%
filter(broad_style %in% top_styles$broad_style) %>%
ggplot(aes(x = score, y = avg, color = broad_style, alpha = 0.4)) +
geom_point() +
theme_classic() +
theme(panel.grid.major.y = element_blank(),
panel.border = element_blank(),
axis.ticks.y = element_blank(),
legend.position = "none") +
facet_wrap(~ broad_style, scales = "free") +
labs(x = "Beer Advocate Score", y = "Avg User Rating",
title = "Comparison Between Beer Advocate Scores and User Ratings")
